What driver should I use 5060ti?

[Hisonor (Jeff Grabenstein)]

I updated to RTX 4070 Ti super a while back and have always tried to have the most recent Nvida Studio drivers.
As long as I do not use the AI presets Photo Lab 9.0.2 works well. I can export 15 images in 35 seconds.
No success with any of the numerous drivers, so I suspect it is hardware, driver, software miscommunication.
Windows 11
Ryzen 9 3900X
64 Gigs of RAM

I do like the auto masking and merging of masks.

[RaceDog]

16GB card here and things are still working great.

[RaceDog]

Try the rebuild option on your newest install and see if there is any change.

[eriepa (WIN11, PL9, FP7, VP4, NIK1)]

@RaceDog … Try the rebuild option on your newest install and see if there is any change.

Thank-you for the follow-up question. Yes, I did try to repair both the previous and current PL9 installs - no luck with either repair. But, I’m fortunate to have a backup PC that I can use in the interim, so will just do another clean install when DxO and / or NVIDIA issue a fix.

[OzarkNerd]

@RaceDog

Just received this card and I still have crashing issues with AI masks. I downloaded and installed the oldest release on their site - 576.02 - still crashes.

I have a 5060 Ti 16GB and I’ve been running 576.88 without a single crash. 64GB system RAM. Windows 10 22H2

EDIT: I saw further down in the thread that you are up and running, that’s great!

[Bytewurm (Kurt O. Wörl)]

As I wrote, the NVIDIA app should show the PL9 application under “Graphics.” Use the “Reset”/“Optimize” button. PL 9 should then no longer be optimized.

[BHAYT (Bryan [WIN10, PL9, VP5, FP7, G9, 5060Ti(16GB), 3060(12GB)])]

@Bytewurm It made no difference to the failed “Sky” attempt on the 5600G with the 3060(12GB) but reverting to an earlier driver (572.83) resolved the issue.

Having reverted to 572.83 and tested, I checked the settings and they were “normal” so I set them to “Optimised” and repeated the test and it was still O.K.

But I have a snapshot of the same test done earlier today with DP3 selected and the memory usage was 12,092MB which is getting close to the memory limit of the 3060(12GB)

[mwsilvers (Mark - New Jersey, USA)]

I also use Nvidia driver 572.83. I changed PhotoLab 9 and PhotoLab 8 in the Nvidia app from Optimized to non optimized and rebooted my system to clear memory before testing the effects of optimization on PhotoLab. 9.

While I did not do any extensive testing, my initial impression is that turning off optimization provided no improvement when using the AI masking presets. When selecting AI masks manually I did not notice any significant difference in performance between optimized and unoptimized PL 9. Manually selecting AI masks worked equally well with PL9 optimized and non-optimized. As a result, I reset the Nvidia app to optimize PhotoLab.

Mark

[BHAYT (Bryan [WIN10, PL9, VP5, FP7, G9, 5060Ti(16GB), 3060(12GB)])]

@mwsilvers

rebooted my system to clear memory before testing the effects of optimization on PhotoLab. 9.

@mwsilvers Provided you can get your PC in and out of ‘Sleep’ successfully that will clear your VRAM, of PhotoLab contents at least. It won’t change anything else, or at least I haven’t noticed that it does.

BUT like other users the failure with the 3060(12GB) was immediate and nothing (short of changing the drivers to an earlier version) would fix it.

Like you I feel that NVidia optimisation has little or no effect but it seems easy enough to make the change and the use ‘Sleep’ and effectively wipe the slate clean i.e. empty VRAM if PhotoLab’s “detritus”.

If you are experiencing no problems with PhotoLab AI when editing but things seem to go wrong during export then

1. Make all your edits and select all the images to be exported
2. Select ‘Sleep’ and when the PC is fully asleep,
3. Wake the computer and execute the export. The export operation will take the minimum amount of VRAM instead of adding the export operation to the VRAM already in use.

[mwsilvers (Mark - New Jersey, USA)]

My card is actually the RTX 4060 with 8 gigabytes. When using the most current Nvidia driver I was having lots of issues with AI masking. A combination of using the older driver and manually selecting the masking rather than using the masking presets, has resolved the problem for me very significantly.

The exports of my Nikon Zf’s 24.5 megapixel RAW files on my older Windows 10 machine, with AI masking applied, is fast. Single images exported using DeepPrime 3 take around 5 to 7 seconds. The same images exported using DeepPrime XD2s take around 8 to 10 seconds.

Every once in a while an export will still take much much longer than it should based on the number of edits. I’ve had exports of DeepPrime 3 take upwards of 25 or 26 seconds and DeepPrime XD2s take up to 42 or 43 seconds for no apparent reason. Note that these are exceptions and not the rule, and if I get out of PhotoLab and go back in and export the same images again they will export very quickly. So clearly, there is still a performance issue with exports but it happens so infrequently for me thet I’ve ceased to worry about it.

Mark

[Evildad]

This thread is interesting because I just had to rebuild my PC (fried mobo and GPU), and I’ve been trying to decide how to replace my RTX3060 12GB. They’re still available at around $250, and it appears that PL wants lots of VRAM, so if I decide to buy a 5060, it should be a 16GB card. Is that pretty accurate? I have not upgraded to PL9.

[Wolfgang ([ Win11, PL9, FP7, VP5, Nik7+3.3 ])]

@Evildad
If you want to spend money you may like to check here and here.

[Stenis (Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))]

I´m not sure at all that it is only about that memory. See below for the vs between RTX 3060 and 3060 Ti. I have the 3060 Ti with 8 Gb (and the 3060 has 12 GB) and it runs perfectly fine if I just stay away from the premade AI-presets and instead use the “Add a Selection” and “Add an Area” like Mike already has told you.

Here is the VS. and an explanation

[BHAYT (Bryan [WIN10, PL9, VP5, FP7, G9, 5060Ti(16GB), 3060(12GB)])]

@Stenis

I have the 3060 Ti with 8 Gb (and the 3060 has 12 GB) and it runs perfectly fine if I just stay away from the premade AI-presets and instead use the “Add a Selection” and “Add an Area” like Mike already has told you.

@Stenis and you believe that you are “secure” for the future? Plus you have defined the limits inside which you need to stay if you want to “guarantee” that your system won’t fall over for lack of GPU memory.

Your 3060TI(8GB) is certainly faster than my 3060(12GB) but right now I wouldn’t swap my GPU for yours, but then I now have a 5060TI(16GB) in my other machine anyway.

@Evildad

so if I decide to buy a 5060, it should be a 16GB card. Is that pretty accurate?

@Evildad I have to be careful not to spend someone else’s money and I would not recommend anyone rushing out to upgrade their GPU, not least because work undertaken by DxO might reduce the memory requirements, somewhat and make 8GB cards look more secure and not with any limitations applied to what the user can do.

I have written here and elsewhere that using “Sleep” to clear GPU memory, has allowed me to set up an export run and then “Sleep” the machine and after the PC wakes, I export.

But I have also commented on the fact that as soon as I then select an image, or change the export option, the memory usage jumps up a lot.

I believe I have now discovered the reason for the sudden jump when I make a single edit, it is because these two options were set

Resetting those options results in no sudden jump unless AI has been applied when having exported with a lower memory usage after a “Sleep” , the memory usage jumps as soon as any edit operation is made.

Using the Loupe is the equivalent to having the options set, i.e. it also causes the memory usage to jump, and I believe it is the second option that is causing the “problem”, but I have had enough of testing for the moment to verify that guess!

So if you want to use all the AI options, including AI presets like “Sky”, with the preview options set and invoke the Loupe and export images, then my current advice it to delay any purchase until we seem to have a more stable release and/or consider cards of 12GB or 16GB to “hedge” your bets but only if you can afford it.

In my case when I discovered my 2060(6GB) was simply not big enough any more, I moved my 3060(12GB) to the 5600G, replacing the 2060(6GB), and looked for what I could afford with 12GB or 16GB for the 5900X, considering another new or second hand 3060(12GB) as an option.

While I was looking, an offer came up on Amazon for a 5060Ti(16GB) at £369.

That was £30 more than I had paid for the 3060(12GB) at the beginning of 2023 and I decided it was too good to miss and am generally pleased with the purchase, the 5900X is twice the speed of the 5600G and the 5060Ti(16GB) is slightly less that twice the speed of the 3060(12GB) and I am hoping both are big enough to cope with the current PL9 release, now and when “fixed” and anything else that DxO might add.

I cannot guarantee that my tests are accurate but in some tests the 12GB card seems to be almost full and I have done other tests where the 16GB card is showing memory usage over 15GB!?

It is possible that DxO are simply using as much space as is available to minimise VRAM memory management but … !?

“You pays you money and takes your choice”, and remember that (currently) “Sleep” clears the VRAM and edits start to fill it again, particularly if you have the ‘Preview’ options set, use the Loupe and definitely if you use AI at the same time.

[BHAYT (Bryan [WIN10, PL9, VP5, FP7, G9, 5060Ti(16GB), 3060(12GB)])]

I apologise if the above post seems a little “extreme” so here is one full of snapshots.

The background to this test group of images is that I went looking for some X-Trans images so that I could test to see how effective XD3-XT was at denoising.

I also wound up downloading a number of Bayer images of Fuji origin (by accident), which I segregated to a separate directory. So I took one of those (Bayer) images that looked like an ideal candidate for a “Sky” test and did a test with the Preview options off and it went as follows, on my 5600G with 3060(12GB), sorry this is essentially all pictures, i.e. the “evidence” that prompted me to write the previous post.

But first I showed one of my images that failed with the 3060(12GB) but had succeeded with the 5060Ti(16GB)(!?), which then worked with the earlier drivers in an earlier post in this topic,

Having “discovered” the “issue” with the ‘Preview’ options I reset those options and installed the latest drivers (put the PC to “Sleep” and restarted) and retested the “Sky” issue and it failed immediately, so not all problems are memory related!

PL902 after a power on and an XD2s export:-

PL902 after the Loupe:-

PL902 after AI hovering to select the sky:-

PL902 using “Sky” AI selection:-

PL902 after export of the single image with the “Sky” mask and LA Clearview set to max:-

Setting Clearview to full to give PL902 some actual work to do with the mask did not increase the memory use so I have omitted one snapshot in the sequence.

[sloweddie (Ed)]

@mwsilvers

@BHAYT

@Bytewurm You have me puzzled by this statement, where is this option you are referring to!?

I think he may be referring to Windows application performance optimization which if I recall correctly can be modifiedn on an application by application basis. I don’t recall the settings offhand but I think among other things it controls background and foreground application processing.

Maybe this (from Windows 11 Settings):

[Wlodek]

@BHAYT

and it failed immediately

@BHAYT, did you also get nvlddmkm event 153 in Win System EventLog?

[BHAYT (Bryan [WIN10, PL9, VP5, FP7, G9, 5060Ti(16GB), 3060(12GB)])]

@Wlodek

@BHAYT, did you also get nvlddmkm event 153 in Win System EventLog?

@Wlodek Sorry I hadn’t checked, so I just did

Certainly one of them coincides with the event in the log, others might but it is dinner time!

What is the significance of this?!

[Wlodek]

@BHAYT

What is the significance of this?!

It’s a kind of generic NVIDIA kernel driver error, perhaps something to do with TDR (Timeout detection and recovery) in Microsoft WDDM. It may happen for example if NVIDIA driver timeouts on a disk I/O while using its cache, but there might be multiple other, very different reasons.
It seems that PL9 AI masks are using some unstable code path, which probably includes code from DxO, NVIDIA, and Microsoft. Interactions could be hard to track. Anyway, it’s time to wait.
BTW, I didn’t get any errors from nvlddmkm with PL8, or PL7, or C1 16.x, or Affinity 2.6.x.

EDIT:

1. In my case, the 153 event has this text “Error occurred on GPUID: 100”. Maybe 100 is CUDA error code ‘cudaErrorNoDevice’ (??)
2. During PL9 exports I also got this in Application EventLog:
   Source: Application Error, Event ID: 1000, Level: Error, Task Category: Application Crashing Events.
   Faulting application name: DxO.PhotoLab.ProcessingCore.exe, version: 9.0.20348.68, time stamp: 0xe166dca2
   Faulting module name: D3D12Core.dll, version: 1.614.1.0, time stamp: 0x66974802
   Exception code: 0xc0000409
   Fault offset: 0x00000000000aa82d
   Faulting application path: C:\Program Files\DxO\DxO PhotoLab 9\DxO.PhotoLab.ProcessingCore.exe
   Faulting module path: C:\Program Files\DxO\DxO PhotoLab 9\D3D12Core.dll

Strangely enough, exports finished successfully, with no visible errors.
The same error with the very same fault offset occurred also for some other applications to other people. No trusty explanation found, Google AI mentions OpenCL compiler issue, but it’s wrong so often, just propagating faulty diagnostics…